On Mon, 21 Feb 2000, Chris Fowler wrote:
> I'm creating a diskless workstation with RH6.1.
>
> Can I just copy everything that is installed to a /ws directory and export
> that to the workstation as rw and ns_root_squash. Then the kernel on the
I'll second both Vernard and Chris on the suggestion of reading
through the "diskless HOWTO". You'll find it quite informative.
You'll find that the diskless box needs a few directories of it's
own. Primarily, /etc. I think in the setup that I wound up using, I shared
/usr and /home between my diskless "client" box and "server" box, with
everything else being a copy.
You should find that the hardest part of getting a diskless box up
is in getting the server set up. You've got to enable tftpd (un-comment
the entry in /etc/inetd.conf). I personally used the bootp/tftpd setup
which was suggested in the HOWTO at the time, though, I do not think that
tftp is the safest thing in the world, so, take caution if you're on an
insecure network. Either way, on the server, you'll have to enable tftpd
and bootpd, and then hack /etc/bootptab. From what I can recall, bootptab
was very picky in it's setup, but, at least on my RH 6.1 box, it's got a
man page which should be able to help you through. I don't remember there
needing to be any major kernel changes on the server side.
One other side point, if you don't know how to "follow"changes
being added to /var/log/messages and /var/log/secure, then practice it a
few times before delving into the diskless arena. You'll need it (or at
least, I did). Take a look at the "tail" utility, or "less" works too
(Press "F" while lessing the file to enter "follow" mode, Control-C to get
back into normal "less" mode).
For the client, you'll have to build a new kernel. In the kernel,
you'll have to remember a few fundamental points:
* Build your net card drivers *into* the kernel, not as a module.
* Select "root filesystem on NFS" before looking for the "get IP
address from bootp / rarp server".
The first point is simple, but, can catch you if your're not
careful. The second point, I should explain. There are two basic parts of
the diskless client which the kernel setup will affect. The first tells
the diskless client how to find it's server (bootp or rarp), the second
tells the kernel not to look on a local disk for it's root filesystem. The
caveat is that, at least when I ran a diskless setup, you'd never see the
options for "bootp/rarp" show up until you selected that you wanted "root
filesystem on NFS" (I may have gotten the order of these flipped). This
became particularly frustrating because the "root filesystem on NFS"
option lies under the "filesystems" kernel config, whereas the
"bootp/rarp" configs were under "Networking" if I remember right.
Another client side kernel related caveat which I remember having
run into was the "/dev/boot255" problem. Basically, you have to mknod a
device, and then use "rdev" to "tell" your client boot disk that it's root
partition is actually an NFS mount, instead of whatever the root partition
currently is. Take a look at linux/Documentation/nfsroot.txt, and search
for "boot255 c 0 255".
I hope all goes well for you, and post if you have any problems.
Most of all, keep chugging away at it. It took me a few very lates nights
to get my diskless setup running, but it was quite worth it once all was
well.
- Mike
====================================================================
Michael Kachline CS, Georgia Institute of Technlology
">kachline@brightstar.gt.ed.net
http://brightstar.gt.ed.net/kachline
====================================================================
--
To unsubscribe: mail ">majordomo@ale.org with "unsubscribe ale" in message body.